Unverified50% confidenceFactExact time
NVIDIA的k8s-device-plugin会将每块物理GPU注册为不可分割的资源单元（nvidia.com/gpu: 1），调度器以此为原子单位分配
1
Sources
50%
Confidence
Long-term
Relevance
7/16/2026
First Seen
Sources
Related Claims
UnverifiedNVIDIA Device Plugin 向 Kubernetes 调度器注册 nvidia.com/gpu 资源类型，使 Pod 可通过 resources.limits 声明 GPU 需求79% similarPartially VerifiedNVIDIA的MIG技术可在A100/H100等高端GPU上实现硬件级别的资源隔离76% similarUnverifiedKubernetes 默认将 GPU 视为不可分割的扩展资源，一个 Pod 请求 1 块 GPU 即独占其全部算力和显存75% similarUnverifiedNVIDIA GPU从Turing架构开始引入INT8张量核心，Ampere架构进一步增强了整数运算能力75% similarUnverifiedNVIDIA GB200 NVL72在单个机架内集成了72颗Blackwell GPU74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/527691API
curl https://kongchang.com/api/v1/knowledge/claims/527691MCP
get_claim(id=527691)